See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/233694427

Finding the eigenmodes of two-dimensional cavities with two axes of

symmetry

Article  in  Acta Acustica united with Acustica · November 1996

CITATIONS READS

4 251

2 authors:

Noam Amir Rudolf Starobinski

Tel Aviv University Silencer. Consulting and Engineering
106 PUBLICATIONS   2,431 CITATIONS    111 PUBLICATIONS   295 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Voice and Emotions View project

no project View project

All content following this page was uploaded by Rudolf Starobinski on 12 April 2020.

The user has requested enhancement of the downloaded file.

 ACUSTICA . acta acustica
Vol.82 (1996)811-823 @ S. Hirzel Verlag . EAA 811

Finding the eigenmodes of two-dimensional cavities with two axes of

symmetry

N. Amir, R. Starobinski
Laboratoire d'Acoustique, URA 1101 CNRS, 8.P.535, Avenue Oliver Messiaen,72}l7 Le Mans Cedex. France

Summary
A method for calculating the modes of vibration of two-dimensional cavities is presented. This method can be used for
shapes that are simply connected, having two axes of symmetry. It is based on a method for computing wave propagation
in waveguides of arbitrarily changing cross section, originally proposed by Roure. The cavity undir coniideraiion is
approximated by a series of rectangles along one of its axes of symmetry, and treated as a waveguide. The input
impedance of this waveguide is used to calculate the resonant frequencies, and calculation of pressure along this
waveguide shows the coresponding vibrational modes. To demonstrate the validity ofthe proposed method, we an-alyze
a circular cavity, with good agreement to analytical results. We then treat a practical case of a semi-elliptical form used
in the cross section of a car mufflers. We also show how this method can be applied to calculate the vibrational modes
of an ideal membrane.
PACS no. 43.20.Ks. 43.20.Mv

1. Introduction ary. If we make an analogy between acoustic pressure in our

calculation and displacement in the case of an ideal mem-
Calculating the acoustical eigenmodes of an arbitrarily brane, we can also calculate the normal modes of such a
shaped bidimensional cavity can be a difficult affair. While membrane, though the boundary condition must be changed.
the analytical solutions for simple shapes like circle, square,
a In a fixed boundary membrane we must use a boundary con-
rectangle and ellipse are known, a numerical solution must dition of zero displacement at the boundary. In the acoustic
usually be implemented for more complex shapes. case this is equivalent to an ideal soft wall.
Various methods appearing in the literature for calculat- One application of the method proposed here is for finding
ing the vibrational modes of ideal membranes could be used the transverse eigenmodes in pipes whose cross section is the
for this problem, such as the Finite Element method (Lerch, form being treated. As an example we will examine a muffler
1991) for example, or more specialized methods as in (Na- with a semi-elliptical cross section.
gaya, 1978:' lne et al., 1981; Cummings and Chang, 1986). The structure of this paper is as follows: we first present
We present here yet another method, pioneered in (Alfred- the basics of the DS method. We then show how it can be
son,1972; Roure, 1976). It is also a rather general method, applied to finding the resonance frequencies and vibration
in the sense that it was originally suggested as a method modes of a two-dimensional cavity. This is followed by a
for calculating wave propagation in acoustic waveguides of short discussion of the accuracy of the method and how it
varying cross section. This method is based on discretizing a can be applied to the analysis of membranes. We suggest a
waveguide along its longitudinal axis, and then performing number of variants on the basic method: in order to evaluate
a computation that "propagates" down the waveguide. Al- them we consider a simple circular form, whose analytical
fredsons analysis is somewhat unwieldy, in that it performs solution is well known. We then apply our method to a semi-
the calculation directly on the pressure field, necessitating elliptical shapes, obtaining further validation.
an iterative approach. Roure showed that "propagating" the
impedance from the load down to the source enables the
computation to be performed with one simple iteration, and 2. Description of the DS method
with no a priori assumptions as to the character of the source.
Using multiple transverse modes, along with a generalized General lossless wave propagation in adiabatic isotropic me-
impedance in matrix form, a high degree of accuracy can be dia is governed by two first order equations, expressing the
achieved with this method. In the sequel we will call this the conservation of mass:
DS (Discrete Segment) method. This problem is very similar
, luJ
to the calculation of the vibration modes of an ideal mem- orv (U) = .-P (1)
brane - the difference being in the boundary conditions for Poc'
pressure in the acoustic cavity case, or the vertical displace- and the equation of momentum conservation:
ment in the membrane case.
The boundary condition used here, based on a hard wall . 1_
Jea = -Vp (2)
condition for fluids, implies a pressure antinode at the bound- Po

u is particle velocity and p is acoustic pressure with the time

Received 25 July 1995, dependance exp(-jo,'t) omitted, the speed of sound is c, and
accepted 22 January 1 996. ps is the density ofair.
 ACUSTICA . acta acustica
812 Amir and Starobinski: Eigenmodes of two-dimensional cavities Vol. 82 (1996)

Figure 1. Arbitrarily shaped waveguide, symmetrical along its lon- Figure 2. A discontinuity between two parallel waveguide sections.
gitudinal axis.

Considering an arbitrarily shaped waveguide, symmetric convenient, as described in (Starobinsky, 1980; Kergomard,
along its longitudinal (z) axis as in Figure 1, we now express 1991). Following a matricial terminology we will write the
the pressure and axial velocity uz using infinite series: above series as:

p:t{i(y,z)Pi(z) (3) P:"!P (9)

i 1

1- u.-5: 1t tbu
r'- - - \ ,''(g,z)u;(z)
h(z) ? where P, U, and ry' are column vectors, and the left super-
script t denotes matrix transpose.
where P1 and Ui are scalar coefficients, h is the half height
Consider now a discontinuity between two rectangular
along the z axis of the waveguide, andrf.;i xe the eigenfunc-
waveguide sections, as in Figure 2. Ifthe pressure and axial
tions obeying the transverse laplacian (41) eigen problem:
velocity on one side of the discontinuity are known, their
values on the other side are obtained by projection over the
ALtbt: -r?rbt (4)
transverse eigenfunctions (Kergomard, 1991) on either side
with the following hard wall boundary conditions: of the discontinuity; pr : p2 and u"1 : L)22 ott ^9r and
u"z : 0 on ^92- ,Sr, thus projections yield:
Etl:,
*:0onC
dy
(5)
Pr : FPz (10 )

and the orthogonality relation: (Jz : tF(Jt

I where the matrix F is defined by:

I ,borb:dS =2h6u
JS
(6)

F',:- 1f
." 2hr I t"1;tlt2.1 dS (1 1)
where ,S is the (one dimensional) cross section, C is the cross Js,,
section boundary, and y is the transverse coordinate. It is
convenient to divide the eigenfunctions into two types: those
In the electrical circuit analogy, the discontinuity is anal-
that are symmetric with respect to the longitudinal axis, and ogous to a multipole transformer (Starobinsky, 1980). It is
those that are antisymmetric. The symmetric eigenfunctions interesting to note that there is no coupling between symmet-
tl,t'" and their eigenvalues afl, as a function of half-height h ric and antisymmetric modes: iooking at (11), it is obvious
afe: that if thu is symmetric and {2i is antisymmetric, or vice
versa, the integral is zero. It is therefore useful to treat the
1, atn:Q, if n -0 (7)
two cases separately. Thus we will have two f'matrices; one
1p; = t/2cosT/, o"*: ff, rf n >U for symmetric eigenfunctions and one for the antisymmetric
ones. For the symmetric case we have:
The antisymmetric eigenfuncttons tl,;fl and their eigenvalues
afl, as a function of half-height h are: 1m: n:0
1

t/2sinc (mra) 1n:}rm ) 0

. (2n-t I)r nm
ltfr: {2sinPt+P, --fL (8) 1
2h,
.-{ 'h, hz (12)
'\2
Note that the trarsverse eigenfunctions are not orthonormal,
as they often are in the literature (Stevenson, 1951); their
2(-I)"--+0'L ,
inner product is always 2h. This can often prove to be more
lma)- - ln)'
xsinc (mn'o) ;otherwise
ACUSTICA . acta acustica
Vol.82 (1996) Amir and Starobinski: Eigenmodes of two-dimensional cavities 813

with o : htlhr, and where sinc(z) : sin(r)/r. In the

antisymmetric case it is:

1-
2n,+1 2m -17
' 2ht 2hz
(- 1)' 2m*1
Ft-: rh1h2
( ) -ff#)
(13)

t"os(/2m+1.h,, \
Z hr"),orherwise
The DS model treats the waveguide as a series of rectan-
gularly shaped segments which approximate the true shape
of the waveguide. A full treatment of this model has been
presented in (Roure, 1976; Roure, 1911) and in (Kergomard Figure 3. P andU at a discontinuity.
et al.,1989); here we give only a short outline.
Dealing directly with the vectors P andU can often result
in numerical problems. It is therefore much more useful to
z(o\ z()
define a generalized impedance, the matrix Z fuIfilling:

P: ZU (14)

Looking at Figure 3, the relationship between the impedance

at the right of the junction 2Q) ,*5is11is the input impedance
to the right waveguide segment, and Z(1), which is the
impedance at the left of the discontinuity is obtained by
substituting (14) into (10):

7(t)ryQ : p7Q)11Q) (15)

tpryG) : UiQ)
Figure 4.P andU at a discontinuity, where the height on the left is
From this we then find, since (JG) be arbitrary: more than on the right.
"un
7G) - FZQ)tF. (16)
transverse modes), ki (k2: - *7)'/'
so that jk.; is real
In the case where the segment on the left is larger than the positive. It can then be easily shown that:
segment on the right, such as in Figure 4, we could use inver-
p(o)
sions of -F- and tF in (16) to obtain Z0) : F-I ZQ)t F-I . - DrpO) +D2ZcLtI) (1e)
We have found it numerically more tractable to use the
same relationships between pressure and velocity to prop-
U(o) : D2Z;1pQ)+DrU1). (20)
agate the admittance matrix across the discontinuity instead Once more, substituting (14) into (19), we obtain:
of Z (Starobinsky, 1981; Pagneux et al.,1995). This matrix
is deflned by g@)
- (nrZttl i D2Z")(D2Z;1Zl) 1Dr)-t. (21)

A similar relationship can be found between y(t) a16 y(z)'

U:YP (17)
y(o) - (DrZ"1 + nrV(r)11pr -t DzZ"y$)1-t. Q2)
and obeys Y : Z-r. Substituting (17) into (10) gives:
Though these equations are algebraic, applying them to areal
y(t) - tFyQ) F. (18) case can cause divergence resembling the divergence of a stiff
differential equation. This is because the eiements of D1 and
To obtain the relationship between ZQ) and the impedance D2 contain very large numbers for the evanescent modes,
at the left of the straight segment 7(o) gee Figures 3, 4), we which cause expontential divergence of the computafion due
define the diagonal matrices Dt, Dz and Z" as follows (index to the finite precision of the computer. Another form which
z refers to the ii'th element of the diagonal): D1 : cos(k1l), is more useful is suggested by Roure for propagating the
Dz : jsin(,kil) and Z,: #.1is the segment length, impedance. Defining Dz: j tan(k1l), we have:
and ki is defined as follows: deflning k : a lc, if k ) a,i
(which is the case for propagating transverse modes), k1 : 2(o) : D;t Z" - D;t Z" (23)
(k2 - a!)t/z; if k < a1 (which is the case for evanescent x lz(1) + D;\ z"]-L D;, z"
 ACUSTICA . acta acustica
814 Amir and Starobinski: Eigenmodes of two-dimensional cavities Vol. 82 (1996)

Similarly, for the admittance we can obtain:

y(o) : Drt z;t -

D;t z;t (24)
x [r(t) + D;r Z;rl-'D;t Z;t
Ifwe are now given a discretized waveguide, and its radiation
impedance, we can propagate this impedance down to the
source, using alternately (16) and(21). Propagating the P or
U down the waveguide is then possible. We can either use
(19) or equivalent equations obtained by substituting (14)
into (19), such as:

(J0) : eD2Z;re1o\ - Z") a

"-i*r1UQ) eS)
To summarize the algorithm for calculating the input
impedance:
1. Discretize the cavity shape
2. Determine the load impedance
3. for n : 1 to (number of segments):
a) propagate Z fromright of segment to left using (23) or Figure 5. A discretized circular cavity
propagate Y from right to left using (24).
b) if next width is bigger, use Y to traverse discontinuity
using (18)
c) if next width is smaller, use Z to traverse discontinuity
using (16)
d) store Zn
After this is finished and Zn are stored, the source condi-
tion must be specified. This can be either by giving P or U .
When one is given, the other ca n be found using the input
impedance, and (25) can be used to propagate pressure and
velocity up to the load.

2. l. Symmetric and antisymmetric transverse

eigenfunction

We will now discuss some of the implications of the fact that

we chose to limit the discussion to shapes having two axes Figure 6. A system for exciting the cavity with propagating antisym-
of symmetry. We assume here that the analysis is carried out metric transverse eigen functions
along one ofthese axes of syrffnetry; the role ofthe second
axis of symmetry will be discussed later.
As shown above the set of transversal eigenfunctions ry'1 frequencies and nodal lines. To avoid confusing these with
can be divided into symmetric and antisymmetric functions, the transverse eigenfunctionsty' we will call the latter "eigen-
and no coupling occurs between them. This can bring a sub- functions", and the former "eigenmodes" or simply "modes".
stantial saving in calculation time, since instead of using a Suppose we now wish to calculate the eigenmodes of a
2n x 2n matrix for ,F', we use two n x n matrices, thus cavity. The example we will deal with throughout this section
reducing the number of elements by half. will be a simple circular cavity. We will divide the problem
On the other hand, while the symmetric set of transverse into two phases: finding the eigen frequencies, and then find-
eigenfunctions contains the plane wave mode, which is never ing the pressure map in the cavity, from which we will also
evanescent, the antisymmetric eigenfunctions all have lower be interested in flnding the nodal lines.
cutoff frequencies. In the following section we will see that
this can be a problem, and we show how to overcome it.
3.1. Finding the resonance frequencies

The first step is to discretize the circle as in Figure 5, treating

3. Applying the DS method to finding the eigenmodes one end as the "source" and the other as the "load". In order
of a cavity to propagate the impedance from the load to the source, we
must then deterrnine the load impedance. In accordance with
In this section we wish to show how the DS method can be the hard wall boundary condition, The conect load should be
used to find the eigenmodes of the cavity, i.e. its resonance a totally reflecting one. This load translates to an impedance
ACUSTICA . acta acustica
Vol. 82 (1996) Amir and Starobinski: Eigenmodes of two-dimensional cavities 815

matrix with infinity on the diagonal, or a zero admittance (non-evanescent) eigenfunctions. the reflected wave (Pag-
matrix. It provides no energy dissipation, therefore implying nuex e/ al., 1995) rs
infinite peaks in the impedance along the cavity. This can
cause numerical problems, therefore it may be conducive to tf -: (2"+ Zi,)-l(2, - zin)u+ (26)
introduce a miniscule dissipation, for instance by using a
very narrow infinite pipe as a load. This works only for the and the energy reflection coefficient is then (see Pagneux
symmetric transverse eigenfunctions, where the plane wave et a1.,1995):
mode is never evanescent. For the antisymmetric transverse
eigenfunctions a load that is always dissipative is harder to D Ute{/u z,u -} (27)
create. One possibility suggested in (Pagneux et al.,1994) fte{t(J 1Z"U a}
is to treat a system as shown in Figure 6. Tubes #1 and we can perform a frequency sweep and search for local
#4 in this figure are wide enough over the frequency range minima. This method was originally suggested in (Pag-
in question so that there is always at least one propagating neux, 1994) for antisymmetrical transverse eigenfunc-
mode in each. On the other hand tubes #2 and #3 are very
tions, and should therefore work for both the symmetric
naffow; the "tunneling" effect at resonance, as described in
and antisymmetric cases, but only in cases in which there
(Pagneux, 1 994) enables energy to propagate from the source
is some form of energy loss.
to infinite tube #4 even though all transverse eigenfunctions
4. One more possible method we suggest here utilizes the
are evanescent in tubes #2 and #3.
second a,ris of symmetry, perpendicular to the flrst, and
Depending on the case, the lossless or lossy models have located at its center; It is only for this method that this
their advantages. In the lossless model all admittances and second axis of syrnmetry is necessary in this paper. The
impedances are purely imaginary, which makes numerical existance of such additional symmetry dictates that each
calculations faster. This model cannot be used in one of the eigenmode of the cavity will have either a minimum or
schemes proposed below, on the other hand, besides being maximum of pressure on this axis. These can be found
potentially diverging. It is important to determine whether by performing a frequency sweep of impedance and ad-
the resonance frequencies obtained in both models are the mittance at this axis, and using one of two methods: the
same; we will examine below both possibilities, showing in first entails searching for maxima tn Zin(0,0) (indicar
fact that the dissipation has very little effect on the accuracy ing pressure maxima) and dr(0,0) (indicating pressure
of the results. minima), following method 2; the second is to search for
Once the impedance has been propagated to the source, it minima in the absolute values of det{ Zin} and det{Yi-},
must be used to determine the resonance frequencies. In the following method 1. In general, as we show below this
simple plane wave case, where Z is a scalar, the resonance method is necessary only for calculation using the anti-
frequencies occur at the peaks of lZl when it is plotted vs. symmetrical eigenfunctions. Since in the circle all of the
frequency. Inott case Z is a matrix, which makes things more azimuthal modes can be considered as either syrnmetri-
complicated. Therefore a number of schemes is possible: cal or antisymmetrical, depending on the orientation of
the nodal lines, for this case we will be able to compare
1. Roure originally pointed out that at resonance the veloc-
the present method to the previous ones, in order to ex-
ity at the volume source is zero; since the pressure is not
amine its validity. For more complex shapes, such as the
thus constrained, the relationship 0 : :
U YP suggests semi-ellipse treated later, use of the symmetric and anti-
that at resonance the matrix Y is singular. In practice, we
symmetric eigenfunctions will give different sets of modes
would perform a frequency sweep, calculating the abso-
and conesponding modal frequencies.
lute value of the determinant of Y at the source for each
frequency, looking for local minima. We will compare these methods in a later section. Before
2. Since the source is placed in a very naffow tube, most if that, we will also discuss how to calculate the pressure in the
not all of the higher order eigenfunctions will be evanes- cavity for a given frequency, in order to flnd the nodal lines.
cent there. Therefore the dominant term in Z, when using
symmetric transverse eigenfunctions, will be the corner
term Zin(0,0), representing the "autoimpedance" of the
3.2. Calctiating the pressure and finding the nodal lines

plane wave mode, which is never evanescent. In this case it Once the resonance frequencies are known, we can excite the
might be sufficient to search for local maxima in this term cavity at each of these frequencies and calculate the pressure
alone. When using antisymmetric transverse eigenfunc- fleld inside. Plotting the pressure magnitude in shades of gray
tions. which are all evanescent there. this method fails. as over the surface of the circle (henceforth "pressure magnitude
shown below. plots") will show the nodal lines as "troughs" in this plot. This
3. Using a velocity source, the greatest pressure will be cre- process is relatively straightforward, though certain care must
ated at the load at resonance. This will create maxima be taken to obtain good results.
in the energy transfer, or minima in the energy reflection For the symmetric modes, the excitation can be a rigid
coefficient. To find the energy reflection coefficient, we piston in the cavity wall. As shown in Figure 5 discretization
assume that the excitation is a rightgoing velocity wave creates a very small height for the source segment, so that
U1, which has non-zero elements only for propagating the solution is not very sensitive to the deviation from a
 ACUSTICA . acta acustica
816 Amir and Starobinski: Eigenmodes of two-dimensional cavities Vol.82 (1996)

perfect hard boundary. The mode excited by such a piston in

the first segment is the plane wave mode, which is always
propagating. From the point of view of the cavity this is
actually almost a point source.
For the antisymmetric eigenfunction, which do not include
the plane wave, any excitation in the narrow source segment
will be of evanescent eigenfunctions, and it remains to be
7v
seen in the following section if this can give a good result.
Propagating the excitation throughout the waveguide has
been briefly discussed in the previous section. Z orY must
be stored at each point in which we wish to calculate the
pressure, and then the value of P or LI is calculated at each
such point, using a formula such as (25). A more detailed
discussion can be found in (Pagneux et al.,1995). Figure 7. A close look at the true contour and the discretized countour
used in the discussion of accuracy

3.3. Applying the method to analysis of a membrane

In the case described above, we are searching for pressure where C1 is the "staircase" contour (see Figure 7). The eigen-
antinodes at the boundary. If we make an analogy between functions $p are notmalized so that (Starobinsky 1980):
pressure for our case, and displacement for a membrane, then
we have to change the boundary condition. In this case it will
be a perfectly soft wall condition; flrst the transverse eigen- lro*öoo, - v6* (32)

functions ry' will have to be changed accordingly, then the where V is the primary area (with smooth contour).
termination should be changed from a positively reflecting
Multiplying (28) by yi and (29)by Sp,we find, by using
one to an acoustic short circuit (zero impedance). Finally,
Greens theorem:
schemes 1-4 above should be "inverted", i.e. we will be look-
ing for peaks in the admittance instead of impedance, and
minima in the absolute value of the impedance determinant - I"rrHot- (Ä; - "il l"sig1,dv:o (33)

instead of the addmitance determinant.

On the contour C the following condition for the small area
AV is satisfied to first order in ÄV:
4. Discussion of accuracy
0u; .'AV Äjhyt
.,.
ü: Ut:
^i
(34)
In this section we attempt to estimate the difference between
the eigenfrequencies of a cavity with smooth walls, and those
"
where AV is the area between the primary and "staircase"
of the discretized cavity used in our analysis, having walls contours, and h is the effective thickness of this volume:
resembling a staircase function. This discussion is valid under h : LVIC, where C here is the length of the smooth
the assumption of using an infinite number of transverse contour. We then have:
eigenfunctions, and therefore the accuracy will be further
limited by truncating their number. In the next section we
will make such a comparison for a circular cavity, to see l.r,ökhd,s
+ Ä?i - "i) Iraiöx<lv :o (35)
^?
whether our estimate holds.
We denote the normalized eigenfunctions of the primary One can decompose 96 in the primary volume V using the
cavity (smooth boundaries) as @6, and the respective eigen- eigenfunctions /7, of the primary area as follows:
values as up,the eigenfunctions ofthe discretized cavity yi
with eigenvalues ,\.7 (see Figure 7) These obey:
ai:ioioöo (36)
Y'fp+uiÖt=0
-t, (28) i'-I
where the coefficients aji arc obtained from (35), without
V2yi t \]a1 :0 (2e) deriving term by term, giving:

with boundary conditions:

(\] - ufl)a1x + m\]irinBon: o (37)

ry-:o
dn,
onc (30)
where m : Vr lV,V1 is the sum of small areas between the
where C is the primary (smooth) contour, and n is the normal, primary and staircase contours, and
and
0ai
En
:0 onCr (31) B,i,*:
i I.r- oihd's (38)
ACUSTICA . acta acustica
Vol. 82 (1996) Amir and Starobinski: Eigenmodes of two-dimensional cavities 817

discussed below where there is no analytical expression this

cannot be ensured.
Another problem influencing the accuracy is that of se-
lecting the number of transverse eigenfunctions to take into
account. From the point ofview ofcircuit theory, this is equiv-
alent to limiting the number of connections in the equivalent
electrical circuit, namely the number of poles in the multi-
pole transformers describing the discontinuities. We will not
deal with this problem in great detail. For the problem of
finding the eigenfrequencies, it is usually sufficient to take
into account the propagating modes plus 2 or 3 evanescent
ones. From the point of view of calculating the pressure fleld,
Figure 8. An alternative discretization of the contoure the considerations are different: supposing we have a certain
pressure distribution at a certain cross section, we need to de-
compose this function over the ly' transverse eigenfunctions
Analogical transformations are true for three dimensional being used. This is purely a geometrical consideration, with
systems as well (Starobinsky, 1983; Herzog and Bruneau, no connection to whether the eigenfunctions being used are
1989). evanescent or not. In general, the more complex the pressure
The system (37) is a linear homogeneous system ofequa- distribution, the more eigenfunctions we need; if we use an
tions forthe a7z coefficients. The deviations ofthe eigenval- insufficient number, this will show up as ripple artifacts in
ues and eigenfunctions between the primary and "staircase" the pressure plots. This point is discussed more in detail in
cavities depend on the small parameter rn. It is easy to show (Pagneux et a\.,1995).
that m - 7 f n, where n is the number of segments in the dis-
cretization. Thus, when the number of segments increases,
the accuracy increases, as a rule. 5. Results for a circular cavity
In order to estimate the order of the error, we make a Taylor
expansion of .\'] and ait inthe small parameter rn. In this section we apply and compare the methods suggested
above to a circular cavity, in order to find the optimal com-
: u? *^rtt) +...,
+*2111") (3e)
bination. The eigenvalues pi of a circular cavity ofradius .R
are the solutions of:
^?j
+rn".tl) +m'a\| +..., (40)
'io: "\f;)
*1,(a,): J,a1lßi) (44)
lJi
*tt".. pjl) u"A *l') are constants. In the absence of degen-
erate modes, i.e. ui f ui,itis easy to obtain: The eigenfrequencies fi obey:

2nh
: -Bii 0t
P!t) (41)
cR- (45)

t\o1,): öi* (42)

Choosing R as cf 2 gives us a dimensionless frequency fi :
.\tn) : Bin (43) Bilr.The flrst ten eigenfrequencies are given in Table I, to
four significant digits.
so the deviations of the pressure distributions of the eigen- One of the advantages in examining the circle is that all
functions and the deviations of the eigenvalues are propor- modes having at least one nodal line on a diameter can be
tional to rn. considered as both symmetric or antisymmetric modes, de-
The problem of estimation of accuracy is more compli- pending on whether a nodal line coincides with the z axis or
cated when degenerate modes are present. As showed in not. We will deal here first with the symmetric modes, and
(Starobinsky, 1983), the difference between the degenerate later with the antisymmetric modes. The verification used for
eigenmodes can be very big, yet the transfer function aver- the accuracy in finding the antisymmetric modes will be done
aged over the frequency band does nat change very much if by examining how well they coincide with the corresponding
rn is small. symmetric ones.
One can increase the accuracy of the approximation fur-
ther by choosing the staircase contour as shown in Figure 8.
5.1. Symmetric modes
If the discontinuity at each step is chosen so that V3 : y',
the Bip are proportional to I f m because of integral (38), We now perform a calculation of Zin on a range of frequen-
and therefore the deviations of the eigenvalues are propor- cies from 0.5 to 2.5 for the circular cavity, using l0 symmet-
tional to m2. However, we have chosen to analyze the more ric eigenfunctions, and discretizing the geometry into 200
pessimistic case above, because a discretization as in Figure segments. Fig 9 shows the results of applying three of the
8 cannot be always assured. Though it can be easy to carry above schemes. A visual inspection of ldet{f,;"} I in part (a)
out for the circle, for shapes such as the semi-elliptic one of this figure shows a rather erratic curve. This is due to the
 ACUSTICA . acta acustica
818 Amir and Starobinski: Eigenmodes of two-dimensional cavities Vol. 82 (1996)

Thble I. First ten resonance frequencies of the cavity, in dimensionless frequency.

Resonance # 2 J 4 5 6 '7 8 9 l0

Frequency 0.5861 0.9122 1.2191 1.3373 1.6926 1.697 2.0421 2.1346 2.2331 2.3877

10 '" 10

o
x o 1
1o'13

v\,
c)
(! l
E (!
0.1
\ t
.q

10 - q)
c
/'\:
:\ i/ \
:5 o
o
o)
10 '"
\ \, ,l \ I
()
ct
X
co
0.01
l. \l\/ \
!
YV]
vl o
a)
!
0.001

I 1il. .l
1o'22 0.0001
b

1 0000 1.02

c)
f
E
c
.q)
o
6)
O.9B
\r-l ^:llD:r
o
l!-
c
o
c E 0.94
o c)
O c)
N_
ä
c)
o.eo
I
C
LU

0.01 0.86

Frequency (Hz) Frequency (Hz)

Figure 9. Results of applying schemes 1-3. (a) scheme I on the whole y matrix, (b) scheme 1 on the 3 by 3 cornermatrix (c) scheme 2 (d)
scheme 3.

very low values in Y for the higher order eigenfunctions. clearly visible. Taking a look at the logarithm ofthe pressure
Thus, limiting the determinant to the 3 by 3 corner matrix amplitude in Figure I I shows the nodal lines much better, as
in (b) shows much better defined dips in its absolute value. very low pressure regions. Observing these flgures, we see
Similarly, in (c) we see very prominent peaks in Zi"(0,0), that the low pressure regions are generally large; but the nodal
and in (d) rather smaller dips in Ä. lines do not always fall exactly at their expected locations.
In Table II we summarize the results for the eigenfre- For cases (a) and (d), the very good symmetry suggests that
quencies, showing the relative difference for each frequency. we have indeed found the correct eigenfrequencies, but that
Schemes 7 and 2 give identical results, which remain the for our discretized geometry they simply do not coincide
same when using either the lossy or lossless model. Scheme with those of a true circle. This is simply demonstrated by
3 can only be applied to the lossy model, on lhe other hand, using the exact 7'th eigenfrequency of a circle, as in Figure
in which it gives errors nearly the same as the above two 12. In this figure the symmetry is gone, and the ratio between
methods, yet missing one of the closely placed frequencies. highest and lowest pressure is lower. Thus, the true resonance
All the errors are on the order of 0.5Vo with respect to the ofthe circle is not in fact the resonance of our geometry.
theoretical values. Cases (b) and (c) in Figures 10 and 11 are slightly asym-
We now turn to calculating the pressure maps for a few of metric, which may indicate that they are not exact eigenfre-
these frequencies. Figure 10 shows the pressure magnitude quencies of our system. the two frequencies in this case are
plots for frequencies #1, #5, #6 and #7. Since the troughs very close, having a relative difference of only 0.237o , which
in this plot are wide and shallow the nodal lines are not makes it difficult to distinguish between them. An attempt to
ACUSTICA . acta acusttca
Vol. 82 (1996) Amir and Starobinski: Eigenmodes of two-dimensional cavities 819

Thble II. First ten resonances in dimensionless frequency, computed using the three schemes, and compared to theory.

Resonance # 2 3 4 5 6 7 8 9 10

Theor. Freq. 0.5861 o.9'722 1.2197 1.3373 1.6926 1.697 2.0421 2.1346 2.2331 2.3811

Scheme 1 0.5899 o.9773 t.2238 1.3441 1 701 I 1 .7051 2.0544 2.1420 2.2407 2.4032
Rel. Diff.[%] -0.648 -0.525 -0.336 -0.508 -0.502 -0.4'71 -0.602 -0.347 -0.340 -0.649

Scheme 2 0.5899 0.9773 1.2238 1.3441 1.701 1 1.705 I 2.0544 2.1420 2.2407 2.4032
Rel. Diff.[7o] -0.648 -0.525 -0.336 -0.508 -0.502 -0.477 -0.602 -0.347 -0.340 -0.649

Scheme 3 0.5902 o.9777 1.2241 1.3444 1.7035 missed 2.0565 2.r414 2.2407 2.4053
Rel. Diff.[7o] -0.700 -0.566 -0.361 -0.531 -0.644 -0.705 -0.319 -0.340 -0.'737

200

150

100

50

100

Figure 10. Pressure magnitude plots for four modes: (a) #1, (b) #5, Figure 1 1. Log pressure magnitude plots for four modes: (a) #'l, (b)
(c) #6, (d) #7. #5, (c) #6, (d) #7.

reflne the calculation of frequency #5 to six decimal places those having one nodal line perpendicular to the horizontal
(1.701 108) did not improve the symmetry of the pressure line (group l), and those which do not (group 2). The latter
graph, indicating that separating between two such closely will have a pressure maximum somewhere on the vertical
placed modes is at the limits of the accuracy of our method. diameter. Denoting the impedance and admittance matrices
We show below how to improve this result. at the center of the circle as Z^i4 ard Y^i6 respectively,
The perfectly rotationally symmetric modes, such as those we will search for: maxima in Y^ia(0,0) for resonances
at frequencies #3 and #9, shown in Figure 13, present no in group 1, maxima in Z^ia(0,0) for resonances in group
problem. - The nodal lines are perfectly circular. 2, minima in d,et{Z*ia} for resonances in group l, and
minima in d,et{Y^ia} for resonances in group 2. As above,
we will take determinants of 3 x 3 submatrices. The results
5.2. Asymmetrical modes
are summarized in Table IIL
If we observe Zi,(0,0) and det{Yi.} in Figure 14 as a We see that both variants give nearly identical results,
function offrequency forthis case, we see that the peaks are which are also very close to the results given by the sym-
very small, especially at lower frequencies. For this reason, metrical eigenfunctions. We have thus demonstrated that this
schemes I and 2 are unreliable in this case. We therefore method gives good results when using the antisyrnmetrical
try the variants suggested in scheme 4 above. Since there transversal eigenfunctions.
is no observable difference between the lossy and lossless The last method we examine is scheme 3, which neces-
case, we choose to use the lossless one, since the purely sitates an extension to the geometry as in Figure 14. Un-
imaginary Z andY matrices are easier to inverl. Taking note fortunately this method gave unreliable results, due to the
of the fact that using the antisymmetric eigenfunctions we fact that most of the eigenfunctions are very far below cutoff
will find only resonance frequencies for which there is a throughout most of the geometry. As a result the reflection
nodal diameter, we will not expect to find resonances #3 and coefficient is very "noisy" and with only very slight dips at
#9. We also divide the remaining resonances into two groups: the resonance frequencies.
 ACUSTIC A acla acustrca
820 Amir and Starobinski: Eigenmodes of two-dimensional cavities Vol.82 (1996)

15

N-
10

E
6
5

o
100 150 200 250 300 350
l

b
.E0
E

!
s 04

100 150 200 250 300 350 4oo

Frequency (Hz)

Figure 12. Linear (a) and log (b) pressure magnitude plots for the Figure 14. Corner value of the Z matrix (a) and absolute value of
theoretical resonance frequency #7. the Y matrix (b) for antisymmetric modes.

200
a b
150 :l

'100
:
il
'a::
.: ..:: :,:

50 #."#*:,"","L=.*-* ,1.

100
20a

150

100 100

50 50
, lll:talt

0 0

100 100

Figure 13. Linear (a,c) and log (b,d) pressure magnitude plots for Figure I 5. Log pressure magnitudepiots for four modes: (a) #1 (b) #5
two circular symmetric modes, #3 (a,b) and #9 (c,d). (c)#6, (d)#'T,calculated with asymmetric transversal eigenfunctions.
Note horizontal nodal line in all cases.

It remains to be seen ifthe pressure can be reliably calcu- 6. Practical example semi elliptical shapes
-
lated with the antisymmetric eigenfunctions. We look once
more at the log pressure magnitude plots of resonances #1, In this section we study a semi elliptical shape taken from a
#5,#6 and#1, in Figure 15, this time calculated using the an- car muffler cross section. The original purpose in analyzing
tisymmetric eigenfrequencies. We observe that all the nodal this shape was to find areas of low pressure in a number of
lines are rotated 90 degrees with respect to the previous calcu- lower order modes, in order to insert a perforated cylindrical
lation, so that the previous vertical nodal line now coincides tube along the muffler, which would excite these modes to
with the horizontal axis. Interestingly, the results are better a minimal degree. There is no analytical description of the
than those obtained with the symmetric eigenfunctions. This cross section - the boundary is given as a series of (X, Y)
is probably due to the fact that the antisymmetric eigenfunc- values along the circumference.
tions "Find" one nodal line (the horizontal) automatically, The study in this section provides a further verification
by virtue of being uniformly zero on the horizontal axis. of our method. We analyze the cavity twice: once along its
It is therefore probably advisable to use the antisymmetric longer axis, and once along the shorter one. We expect to
eigenfunctions wherever applicable. find the same eigenmodes, within a reasonably small enor.
The frequency band considered was between 500 and
4500H2. Using scheme 2 for the symmetric eigenfunctions
and scheme 4 for the antisymmetric eigenfunctions, we de-
tected 11 symmetric modes, and 9 antisymmetric ones. Out
of the latter, 5 are antisymmetric in one orientation and sym-
ACUSTICA . acta acustica
Vol. 82 (1996) Amir and Starobinski: Eigenmodes of two-dimensional cavities 821
Table III. Eight resonances found using antisymmetric eigenfunctions, and compared to theory.

Group Group #1 Group #2

Resonance# 2 5 8 l0 4 6 '7

Theoretical Frequency 0.9722 t.6926 2.1346 2.3877 0.5861 I .JJ /.J 1.6970 2.0421

Max in Y-;4(0,0), Z*.;a(0,0) 0.9758 1.70t9 2.142 2.4007 0.5869 1.3435 1.7005 2.0528
Rel. Diff.[7ol -0.370 -0.549 -0.347 -0.544 -0.1 36 -0.464 -0.206 -0.524

Mtn tn det{Z *a}, det {Y^ro} 0.9758 1.7018 2.142 2.4006 0.s869 1.3435 1.7005 2.0528
Rel. Ditr.[7o] -0.370 -0.544 -0.341 -0.540 -0. r 36 -0.464 -0.206 -0.524

2272H2 2277 Hz 2330 Hz (sym) 2323 Hz (asym)

120
100 100
b b
100
80
60 50 50
40 40
20 0

0
50 100 150 50 100 150 200 50 100 '150 50 100 150 200

2939 Hz 2934 Hz 2878 Hz (asym) 2877 P,z (sym\

20
100 00
c d
80
50 60
40
0 20
0
100 150 2AA '100 150 200

Figure 16. Log pressure magnitude plots for two resonance frequen- Figure I 7. Log pressure magnitude plots for two resonance frequen-
cies of the semi-ellipse, for two orientations. Modes are symmetrical cies of the semi-ellipse, for two orientations. Modes are symmetrical
in both orientations. in one orientation, antisymme trical in another.

metric in the latter, and four are antisymmetric with respect tations. In Figure 16 we show two log pressure magnitude
to both axes. This means that 16 eigenmodes could be de- plots in both orientations, observing that the nodal lines in
tected in one orientation or another simply with the symmet- the two agree very we11., Figure 17 shows two modes that
ric transversal eigenfunctions, but 4 eigenmodes could be are symmetric in one orientation and antisymmetric in the
detected only with the scheme for antysymmetric transverse other: note that in the symmetric case the method can have
eigenfrequencies. The eigenfrequencies of both analyses are a slight difficulty in finding the precise vertical nodal line,
surnmarized in Table IV, along with the relative difference. whereas in the antisymmetric case it is found automatically.
We observe that the relative difference between the reso- This strengthens the point raised above, that wherever possi-
nance frequencies in both orientations is very small, as ex- ble the antisymmetric case should be used to find the nodal
pected. Further, we observe that some of the frequencies in lines.
the above table are quite close: for example 3510H2 and
3580H2 in orientation #2. As we saw before for the circular
cavity, this can create problems in computing the nodal lines. 7. Conclusion
On the other hand, using orientation #l one of these remains
attributed to the symmetric modes, and the other to the an- We have presented here a method for flnding the eigenmodes
tisymmetric ones; in this orientation they do not "interfere" of a two-dimensional cavity, based on a method suggested
with one another, facilitating the computation of the nodal previously (Roure, 1977), involving the use of a transverse
lines. Thus it can be beneficial to analyse any given shape mode decomposition of the pressure and axial velocity. We
using both orientations. have limited the discussion to shapes having two axes of
We also examine a number of modes more closely, show- syfimetry. The longitudinal axis of syrnmetry enables us
ing how the nodal lines are nearly identical for both orien- to separate the problem into antisymmetric and symmetric
 ACUSTICA . acta acusttca
822 Amir and Starobinski: Eigenmodes of two-dimensional cavities Vol.82 (1996)

Table IV. Comparison of resonances of the semi-ellipse in two different orientations. s-symmetrical, a-antisymmetrical along longitudinal axis
only, aa-antisymmetrical along both axes. A1l frequencies are in Hertz.

C)rientation #l C)rientation #2 Rel. Diff. Orientation #l Orientation #2 Rel. Diff.

Freq. Symm. Freq. Symm. Vo Freq, Symm. Freq. Symm. Vo

916 a 911 S -0.498 3293 S 3300 a 0.2t2

t167 S 1169 a 0.126 354r S 3526 a -0.443
1652 S 1649 S -0.205 3577 S 3570 S -0.196
1'712 aa 17 11 aa -0.043 3576 a 3580 S -0.111
2212 s 2277 S 0.213 3848 aa 3855 aa 0.1 69
2312 a 2309 S -0.11 1 4163 S 4155 S -0.1 89
2330 s 2324 a -0.294 411 4 aa 4160 aa -0.340
28'78 a 2878 S -0.013 4288 a 4219 S -0.200
2939 S 2935 S -0.147 431 8 S 4328 S 0.226
2954 aa 2945 aa -o.29'7 4470 S 4470 a 0.001

transverse eigenfunctions problems, since the two do not Kergomard, J. (l 991). Calculation of discontinuities in waveguides
couple. The vertical axis of symmetry is necessary for finding using mode-matching method: an alternative to the scattering
matrix approach. J. Acoustique 4,111-137.
the resonances using the antisymmetric eigenfunctions.
Kergomard, J., Garcia, 4., Tägui, G., and Dalmont, J. P. (1989).
Applying this method to a circle, we show that it is capable Analysis of higher order mode effects in an expansion chamber
of flnding the resonance frequencies to a precision of about using modal theory and equivalent electrical circuits. J. Sound
0.5Vo. Prediction of nodal lines is not perfect in all cases, r'Ib. t29(3),451475.
Lerch, R. (1991). A new method for the calculation of the mechanical
though the method can definitely be used to predict general
behaviour of arbitrarily shaped membranes. Acustica 73, 100-
low pressure areas. Other studies suggest that finite elements 106.
may give better results, though they require dedicated soft- Nagaya, K. ( 1 978). Vibrations and dynamic response of membranes
ware. with arbitrary shape. Journal ofApplied Mechanics 45,153-158.
Pagneux, V., Amir, N., and Kergomard, J. (1995). A study of wave
Finally we demon strated the method on a practical ex- propagation in varying cross section waveguides by modal de-
ample of a semi-elliptical shape, for which there exists no composition - Part I: theory and validation. J. Acoust. Soc. Am.
analytical solution. We observe that very good agreement is submitted.
obtained between analyses along the vertical and horizontal Pagneux, V. and Kergomard, J. (.1995). Modes localisds dans les
guides d'ondes acoustiques ä section variable. Congrös Francais
axis, and show that the two analyses can in fact be comple-
de M öcanique,Strasbourg.
mentary. Roure, A. (1976). Propagation Buidöe, ötude des discontinuitös.
PhD thesis. Universitd Aix-Marseille.
Roure, A. (1977). Propagation du son dans des conduits a sec-
tion continuement variable - application a la determination des
References frequences propres de certains volumes complexes. Euromech.
Starobinsky, R. (1980). Methods of circuit theory in the problems of
Alfredson, R. I. (1972). The propagation of sound in a circular duct internal acoustics of machines. Jet Engine Noise, Central Institute
of continuously varying cross-sectional area. J. Sound Vb. 23, of Aviation Engines, Moscow, 181-210. (in Russian).
433442. Starobinsky, R. (1981). Some mathematical models of acoustical
Cummings, A. and Chang, I. J. (1986). Sound propagation in a systems. preprint manuscript NIINAWOP ROM - D 5 86, Togliatti,
fl at-oval waveguide. J. S ound Vib. 106(I), 35-43. 1op.
Herzog, P. and Bruneau, M. (1989). Shape perturbation and internal Starobinsky, R. (1982). Theory and synthesis of mffiers for the
mode coupling in cavities. J. Ac oust. So c. Am. 86(6), 237'7 -2384. intake and exhaust of internal combustion engines. PhD thesis,
Irie, T., Yamada, G., and Yoda, K. (1981). Free vibration of mem- Togliatti. (in Russian), 300p.
branes and plates with four curved edges. -/. Acoust. Soc. Am. Stevenson, A. F. (1951). Exact and approximate equations for wave
70(4), 1083-i088. propagation in acoustic horns. -/. App. Phys.22(12),1461-1463.

View publication stats